I wouldn't bet on it. In the "preview" round, Watson seemed to come up with its answers more or less instantaneously. (Or at least, before the host finished reading the question.) My guess is the limiting factor is the quality of the underlying dataset and cleverness of the algorithms, not CPU time.